BULLETIN (New Series) OF THE

AMERICAN MATHEMATICAL SOCIETY

Volume 44, Number 4, October 2007, Pages 515539
S 0273-0979(07)01175-5
Article electronically published on June 26, 2007
EULER AND HIS WORK ON INFINITE SERIES
V. S. VARADARAJAN
For the 300
th
anniversary of Leonhard Eulers birth
Table of contents
1. Introduction
2. Zeta values
3. Divergent series
4. Summation formula
5. Concluding remarks
1. Introduction
Leonhard Euler is one of the greatest and most astounding icons in the history
of science. His work, dating back to the early eighteenth century, is still with us,
very much alive and generating intense interest. Like Shakespeare and Mozart,
he has remained fresh and captivating because of his personality as well as his
ideas and achievements in mathematics. The reasons for this phenomenon lie in
his universality, his uniqueness, and the immense output he left behind in papers,
correspondence, diaries, and other memorabilia. Opera Omnia [E], his collected
works and correspondence, is still in the process of completion, close to eighty
volumes and 31,000+ pages and counting. A volume of brief summaries of his
letters runs to several hundred pages. It is hard to comprehend the prodigious
energy and creativity of this man who fueled such a monumental output. Even
more remarkable, and in stark contrast to men like Newton and Gauss, is the sunny
and equable temperament that informed all of his work, his correspondence, and his
interactions with other people, both common and scientic. It was often said of him
that he did mathematics as other people breathed, eortlessly and continuously. It
was also said (by Laplace) that all mathematicians were his students.
It is appropriate in this, the tercentennial year of his birth, to revisit him and
survey his work, its oshoots, and the remarkable vitality of his themes which are
still ourishing, and to immerse ourselves once again in the universe of ideas that
he has created. This is not a task for a single individual, and appropriately enough,
a number of mathematicians are attempting to do this and present a picture of
his work and its modern resonances to the general mathematical community. To
be honest, such a project is Himalayan in its scope, and it is impossible to do full
justice to it. In the following pages I shall try to make a very small contribution
to this project, discussing in a sketchy manner Eulers work on innite series and
its modern outgrowths. My aim is to acquaint the generic mathematician with
Received by the editors April 20, 2007 and, in revised form, April 23, 2007.
2000 Mathematics Subject Classication. Primary 01A50, 40G10, 11M99.
c 2007 American Mathematical Society
Reverts to public domain 28 years from publication
515
516 V. S. VARADARAJAN
some Eulerian themes and point out that some of them are still awaiting complete
understanding. Above all, it is the freedom and imagination with which Euler
operates that are most compelling, and I would hope that the remarks below have
captured at least some of it. For a tribute to this facet of Eulers work, see [C].
The literature on Euler, both personal and mathematical, is huge. The references
given at the end are just a fraction of what is relevant and are in no way intended
to be complete. However, many of the points examined in this article are treated
at much greater length in my book [V], which contains more detailed references.
After the book came out, Professor Pierre Deligne, of the Institute for Advanced
Study, Princeton, wrote to me some letters in which he discussed his views on some
of the themes treated in my book. I have taken the liberty of including here some
of his comments that have enriched my understanding of Eulers work, especially
on innite series. I wish to thank Professor Deligne for his generosity in sharing
his ideas with me and for giving me permission to discuss them here. I also wish
to thank Professor Trond Digernes of the University of Trondheim, Norway, for
helping me with electronic computations concerning some continued fractions that
come up in Eulers work on summing the factorial-like series.
2. Zeta values
Euler must be regarded as the rst master of the theory of innite series. He
created it and was by far its greatest master. Perhaps only Jacobi and Ramanujan
may be regarded as being even close. Before Euler entered the mathematical scene,
innite series had been considered by many mathematicians, going back to very
early times. However there was no systematic theory; people had only very informal
ideas about convergence and divergence. Also most of the series considered had only
positive terms. Archimedes used the geometric series
4
3
= 1 +
1
4
+
1
4
2
+
1
4
3
+. . .
in computing, by what he called the method of exhaustion, the area cut o by a
secant from a parabola. Leibniz, Gregory, and Newton had also considered various
special series, among which the Leibniz evaluation,

4
= 1
1
3
+
1
5

1
7
+. . . ,
was a most striking one. In the fourteenth century people discussed the harmonic
series
1 +
1
2
+
1
3
+. . . ,
and Pietro Mengoli (16251686) seems to have posed the problem of nding the
sum of the series
1 +
1
2
2
+
1
3
2
+ . . . .
This problem generated intense interest, and the Bernoulli brothers, Johann and
Jakob, especially the former, appear to have made eorts to nd the sum. It came
to be known as the Basel problem. But all eorts to solve it had proven useless,
and even an accurate numerical evaluation was extremely dicult because of the
slow decay of the terms. Indeed, since
1
n

1
n + 1
=
1
n(n + 1)
<
1
n
2
<
1
n(n 1)
=
1
n 1

1
n
EULER AND HIS WORK ON INFINITE SERIES 517
we have
1
N + 1
<

n=N+1
1
n
2
<
1
N
,
so that to compute directly the sum with an accuracy of six decimal places would
require taking into account at least a million terms.
Eulers rst attack on the Basel problem already revealed how far ahead of
everyone else he was. Since the terms of the series decreased very slowly, Euler
realized that he had to transform the series into a rapidly convergent one to facilitate
easy numerical computation. He did exactly that. To describe his result, let me
use modern notation (for brevity) and write
(2) = 1 +
1
2
2
+
1
3
2
+. . . .
Then Eulers remarkable formula is
(1) (2) = (log 2)
2
+ 2

n=1
1
n
2
.2
n
with
log 2 =
1
2
+
1
8
+
1
24
+ =

n=1
1
n.2
n
.
The terms in the series are geometric, and the one for log 2 is obtained by taking the
value x =
1
2
in the power series for log(1 x). However formula (1) lies deeper.
Using this he calculated (2) accurately to six places and obtained the value
(2) = 1.644944 . . . .
To derive (1) Euler introduced the power series
x +
x
2
2
2
+
x
3
3
2
+ =

n=1
x
n
n
2
,
which is the generating function of the sequence (1/n
2
). This is an idea of great
signicance for him because, throughout his life, especially when he was attempting
to build a theory of divergent series, he regarded innite series as arising out of
generating functions by evaluation at special values. In this case the function in
question has an integral representation: namely
(2)

n=1
x
n
n
2
= Li
2
(x)
where
Li
2
(x) :=
_
x
0
log(1 t)
t
dt =
_ _
0<t
2
<t
1
<x
dt
1
dt
2
t
1
(1 t
2
)
.
It is the rst appearance of the dilogarithm, a special case of the polylogarithms
which have been studied recently in connection with multizeta values (more about
these later). Clearly
(2) = Li
2
(1).
The integral representation allowed Euler to transform the series as we shall see
now. He obtained the functional equation
(3) Li
2
(x) + Li
2
(1 x) = log xlog(1 x) + Li
2
(1),
518 V. S. VARADARAJAN
which leads, on taking x =
1
2
, to
(2) = (log 2)
2
+ 2

n=1
1
n
2
.2
n
.
The formula (3) is easy to prove. We write
(2) =
_
u
0
log(1 x)
x
dx +
_
1
u
log(1 x)
x
dx.
We then change x to 1 x in the second integral and integrate it by parts to get
(1). More than the specic result, the signicance of Eulers result lies in the fact
that it lifted the entire theory of innite series to a new level and brought new ideas
and themes.
Still Euler was not satised, since he was far from an exact evaluation. Then
suddenly, he had an idea which led him to the goal. In his paper that gave this new
method for the solution of the explicit evaluation he writes excitedly at the begin-
ning: So much work has been done on the series (n) that it seems hardly likely
that anything new about them may still turn up. . . . I, too, in spite of repeated ef-
forts, could achieve nothing more than approximate values for their sums. . . . Now,
however, quite unexpectedly, I have found an elegant formula for (2), depending
on the quadrature of a circle [i.e., upon ] (from Andre Weils translation).
Eulers idea was based on an audacious generalization of Newtons formula for
the sums of powers of the roots of a polynomial to the case when the polynomial
was replaced by a power series. Writing a polynomial in the form
1 s +s
2
s
k
=
_
1
s
a
__
1
s
b
_
. . .
_
1
s
r
_
we have
=
1
a
+
1
b
+ +
1
r
, =
1
ab
+
1
ac
+. . .
and so on. In particular
1
a
2
+
1
b
2
+ +
1
r
2
=
2
2
and more generally
S
3
=
3
3 + 3, S
4
=
4
4
2
+ 4 + 2
2
4
and so on, where
S
k
=
1
a
k
+
1
b
k
+ +
1
r
k
.
Eulers idea was to apply these relations wholesale to the case when the polynomial
is replaced by a power series
1 s +s
2
. . . ,
indeed, to the power series
1 sin s = 1 s +
s
3
6
. . . .
The function 1 sin s has the roots (all roots are double)

2
,

2
,
3
2
,
3
2
,
5
2
,
5
2
, . . . ,
EULER AND HIS WORK ON INFINITE SERIES 519
and so the above formulas give the following. First,
4

_
1
1
3
+
1
5
. . .
_
= 1,
which is Leibnizs result. But now one can keep going and get
8

2
_
1 +
1
3
2
+
1
5
2
+. . .
_
= 1,
which leads at once to
(2) =

2
6
.
One can go on and on, which is what Euler did, calculating (2k) up to 2k = 12.
In particular
(4) = 1 +
1
2
4
+
1
3
4
+ =

4
90
.
The same method can be applied to sin s and leads to the same results.
Euler communicated these (and other) results to his friends (the Bernoullis in
particular), and very soon everyone that mattered knew of Eulers sensational dis-
coveries. He knew that his derivations were open to serious objections, many of
which he himself was aware of. The most important of the objections were the
following: (1) How can one be sure that 1 sin s does not have other roots besides
the ones written? (2) If f(s) is any function to which this method is applied, f(s)
and e
s
f(s) both have the same roots and yet they should lead to dierent formu-
lae. Nevertheless the numerical evaluations bolstered Eulers condence, and he
kept working to achieve a demonstration that would satisfy his critics. It took him
about ten years, but he nally succeeded in obtaining the famous product formula
for sin s:
(4)
sin x
x
=

n=1
_
1
x
2
n
2

2
_
.
Once this formula is established, all the objections disappear, as he himself re-
marked.
The proof of (4) by Euler was beautiful and direct. He wrote
sin x
x
= lim
n
_
1 +
ix
n
_
n

_
1
ix
n
_
n
2ix
and factorized explicitly the polynomials
q
n
(x) :=
_
1 +
ix
n
_
n

_
1
ix
n
_
n
2ix
to get
q
n
(x) =
p

k=1
_
1
x
2
n
2
1 + cos
2k
n
1 cos
2k
n
_
(n = 2p + 1).
The formula (4) is obtained by letting n go to term by term in the product. As
would be natural to expect, Euler does not comment on this passage to the limit; a
modern rigorous argument would add just the observation that the passage to the
520 V. S. VARADARAJAN
limit termwise can be justied by uniform convergence, as can be seen from the
easily established estimate

1 + cos
2k
n
1 cos
2k
n

C
x
2
k
2

2
where C is an absolute constant. The method is applicable to a whole slew of
trigonometric as well as hyperbolic functions and allowed Euler to reach all the
formulae obtained earlier by his questionable use of Newtons theorem. Among
these are
1
sin s
sin
=

n=
_
1
s
2n +
__
1
s
2n +
_
.
For convergence purposes this should be rewritten as
1
sin s
sin
=
_
1
s

n=1
_
1
s
2n +
__
1 +
s
2n
_

n=1
_
1
s
(2n 1)
__
1 +
s
(2n 1) +
_
.
From the product formula (4) one can calculate by Newtons method the values of
(2k) explicitly; there are no problems (as everything in sight is absolutely conver-
gent), and Euler did this. These evaluations, especially the value
(12) =
691
6825 93555

12
,
must have suggested to him that the Bernoulli numbers were lurking around the
corner here, since
B
12
=
691
2730
.
Euler then succeeded in getting a closed formula for all the (2k).
The main idea is to logarithmically dierentiate (4) (as was also observed imme-
diately by Nicholas Bernoulli) to get (replacing x by s)
(5) cot s =
1
s
+

n=1
_
1
n +s

1
n s
_
(0 < s < 1).
The formula is written in such a way that absolute convergence is manifest; Euler
did not bother with such niceties and wrote it as
(6) cot s =

1
s +n
.
It is denitely more convenient to do this, interpreting the sum as a principal value.
We shall do so from now on, omitting the reference to principal values for brevity.
Expressing the cotangent in terms of exponentials leads one to the function
B(s) = B(s) :=
s
e
s
1
1 +
1
2
s =

k=1
B
2k
(2k)!
s
2k
.
EULER AND HIS WORK ON INFINITE SERIES 521
The B
2k
are the Bernoulli numbers, introduced by Jakob Bernoulli many years
before Euler; Euler suggested they be called Bernoulli numbers. For the rst few
we have
B
2
=
1
6
, B
4
=
1
30
, B
6
=
1
42
, B
8
=
1
30
, B
10
=
5
66
, B
12
=
691
2730
.
Then
cot s
1
s
=
2i
2is
B(2is) = 2i

k=1
B
2k
(2is)
2k1
(2k)!
.
Calculating derivatives at s = 0 we get Eulers surpassingly beautiful formula
(7) (2k) =
(1)
k1
2
2k1
B
2k
(2k)!

2k
.
Nowadays it is customary to treat s as a complex variable and establish (5) or
(6) by complex methods, using periodicity and Liouvilles theorem. I think however
that Eulers method is unrivaled in its originality and directness. For a treatment of
these formulae that is very close to Eulers and even more elementary in the sense
that one works entirely over the real eld, see Omar Hijabs very nice book [Hi].
One should also note that the results of Euler may be viewed as the forerunners
of the work of Weierstrass and Jacobi, of innite products with specied zeros and
poles, with sums over lattices in the complex plane replacing sums over integers (
and -functions).
In addition to the zeta values Euler also determined the values
L(2k + 1) = 1
1
3
2k+1
+
1
5
2k+1
. . . .
These are the very rst examples of twisting, namely replacing a series by one where
the coecients are multiplied by a character mod N:

n1
a
n
n
s

n1
a
n
(n)
n
s
where is a character mod N, more generally a function of period N. The transition
from to L corresponds to a character mod 4:
(n) =
_
(1)
(n1)/2
if n is odd
0 if n is even.
I shall talk more about these when I discuss Euler products. The method for the
sums L(2k+1) is the same as for the zeta values and starts with the partial fraction

sin s
=

(1)
n
1
s +n
obtained by logarithmically dierentiating the innite product
1
sin x
sin s
=

n=
_
1
x
2n +s
__
1
x
2n + s
_
at x = 0 and then changing s to s.
It was natural for Euler to explore if the partial fraction expansions
(8)

sin s
=

(1)
n
1
s + n
, cot s =

1
s +n
522 V. S. VARADARAJAN
could be established by other methods. This he did in several beautiful papers, and
his derivations take us through a whole collection of beautiful formulae in integral
calculus, including the entire basic theory of what Legendre would later call the
Eulerian integrals of the rst and second kind, namely, the theory of the beta and
gamma functions.
The starting point of the new derivation is the pair of formulae
_
1
0
x
p1
+x
qp1
1 +x
q
dx =

(1)
n
1
p +nq
(q > p > 0)
_
1
0
x
p1
x
qp1
1 x
q
dx =

1
p +nq
(q > p > 0).
(9)
These are derived by expanding
1
1 x
q
as power series and integrating term by term. One has to be a bit careful in the
second of these formulae since the integrals do not converge separately. It is then
a question of evaluating the integrals directly to obtain the formulae
_
1
0
x
p1
+x
qp1
1 +x
q
dx =

q sin(p/q)
(q > p > 0),
_
1
0
x
p1
x
qp1
1 x
q
dx =
cot(p/q)
q
(q > p > 0).
(10)
We then obtain (7) with s = p/q. For Euler this was sucient; we would add to his
derivation a remark about justifying the continuity of both sides of the formulae in
s.
For proving (10) Euler developed a method based on a beautiful generalization
of the familiar formula (indenite integration)
_
dx
1 +x
2
= arctanx.
Euler obtains for
_
x
0
x
m1
1 +x
2n
dx (2m > n > 0.m, n integers)
the formula
(1)
m1
2n
n

k=1
cos(2k 1)m

2n
log
_
1 + 2xcos(2k 1)

2n
+x
2
_
+
(1)
m1
n
n

k=1
sin(2k 1)m

2n
arctan
xsin(2k 1)

2n
1 +xcos(2k 1)

2n
.
The formula is obtained using partial fractions and the factorization of (1 + x
2n
).
We now let x in this formula. Using the identities (which Euler derived as
EULER AND HIS WORK ON INFINITE SERIES 523
special cases of a whole class of trigonometric identities)
n

k=1
cos
(2k 1)m
2n
= 0
n

k=1
(2k 1) sin
(2k 1)m
2n
=
(1)
m1
n
sin
m
2n
,
we get, with Euler,
_

0
x
m1
1 +x
2n
dx =

2nsin
m
2n
.
We put p = m, q = 2n and rewrite this as
_

0
x
p1
1 +x
q
dx =

q sin
p
q
(q > p > 0).
Here q is even; but if q is odd, the substitution x = y
2
changes the integral to one
with the even integer 2q, and we obtain the above formula for odd q also. Euler
does not stop with this of course; he goes on to evaluate all the integrals of the
form
_

0
x
p1
(1 +x
q
)
k
dx.
In particular he nds
_

0
x
m1
1 2x
n
cos +x
2n
dx =
sin
nm
n
( )
nsin sin
(nm)
n
.
For =

2
this reduces to the previous formula.
The derivation of the second integral in (10) is similar but more complicated since
we have to take into account the fact that the integrals do not converge separately.
It is based on getting a formula for
_
x
0
x
m1
1 x
2n
dx
and we omit the details. The result is
_
1
0
x
m1
x
2nm1
1 x
2n
dx =

2n
cot
m
2n
,
which leads as before to the second formula in (10). It is to be noted that in this
method also the factorization of (1x
2n
) enters decisively, exactly as in his original
proof of the innite product for sinx.
Finally one could also obtain (10) as a consequence of the theory of the gamma
function, using only formulae that were known to Euler. We are used to writing
(s + 1) =
_

0
e
x
x
s
dx,
but Euler always preferred to write it as
[s] = s! =
_
1
0
(log x)
s
dx
and think of it as an interpolation for n! He knew the functional equation
[s] = (s + 1)[s 1]
524 V. S. VARADARAJAN
as well as the formula
(s)(1 s) =

sin s
,
(the corollary)

_
1
2
_
=

,
and the limit formula
(1 + m) = lim
m
1.2. . . . n
(m+ 1)(m+ 2) . . . (m+n)
(n + 1)
m
,
which he would write as
[m] = lim
m
1.2
m
m+ 1
2
1m
.3
m
m+ 2
. . .
n
1m
(n + 1)
m
m+n
.
In fact it is in this form he introduces the Gamma function in one of his early letters
to Goldbach. The derivation of (10) is now a straightforward consequence of the
theory of these integrals. One gets
_
1
0
x
qp1
1 +x
q
dx =
_

1
x
p1
1 +x
q
dx
so that
_
1
0
x
p1
+x
qp1
1 +x
q
dx =
_

0
x
p1
1 +x
q
dx =
1
q
B
_
p
q
, 1
p
q
_
=

q sin(p/q)
.
Once again the treatment of the second integral in (10) is more delicate.
The partial fractions (9) can be dierentiated and specialized to yield explicit
values for many innite series. Euler worked out a whole host of these, with or
without the twisting mentioned earlier. The sums he treated are of the form

nZ
h(n)
(nq +p)
r
where h is a periodic function, and their values are of the form
g
r
where g is a cyclotomic number. The series he obtains are actually Dirichlet series
corresponding to various characters mod q and their variants. Thus, with as the
non-trivial character mod 3, extended to Z by 0,
2
3

3
=

n=1
(1)
n1
(n)
n
,

3

3
=

n=1
(n)
n
,
which he would write as
2
3

3
= 1 +
1
2

1
4

1
5
+
1
7
+
1
8

1
10

1
11
+. . .

3
= 1
1
2
+
1
4

1
5
+
1
7

1
8
+. . . .
Also

2
8

2
=

n=1

8
(n)
n
2

2
6

3
=

n=1

12
(n)
n
2
EULER AND HIS WORK ON INFINITE SERIES 525
where

8
(n) =

+1 if n 1 mod 8
1 if n 3 mod 8
0 if otherwise

12
(n) =

+1 if n 1 mod 12
1 if n 5 mod 12
0 if otherwise,
which he would write as

2
8

2
= 1
1
3
2

1
5
2
+
1
7
2
+. . .

2
6

3
= 1
1
5
2

1
7
2
+
1
11
2
+. . .
and so on.
Multizeta values. Throughout his life Euler tried to determine the zeta values
at odd integers, (3), (5), . . . but was unsuccessful. He obtained many formulae
linking them but was unable to get a breakthrough. Late in his life, almost thirty
years after his discoveries, he wrote a beautiful paper where he introduced what
are now called multizeta values. The double zeta values are nowadays dened as
(a, b) =

m>n>0
1
m
a
n
b
(a, b Z, a 2, b 1).
This is a slight variant of Eulers denition which we write as
E
(a, b), in which he
would sum for m n and write the sum as
1 +
1
2
a
_
1 +
1
2
b
_
+
1
3
a
_
1 +
1
2
b
+
1
3
b
_
+. . .
so that

E
(a, b) = (a, b) +(a +b).
He proved the beautiful relation
(2, 1) = (3)
as well as the more general
(p, 1) +(p 1, 2) + +(2, p 1) = (p + 1)
from which he derived the relations
2(p 1, 1) = (p 1)(p)

2qp2
(q)(p q).
In recent years people have dened the multizeta values by
(s
1
, s
2
, . . . , s
r
) =

n
1
>n
2
>>n
r
>0
1
n
s
1
1
n
s
2
2
. . . n
s
r
r
(s
i
Z, s
1
2, s
i
1).
The Euler identities have been generalized, new identities have been discovered by
Ecalle and others, and considerable progress has been made about the nature of
these numbers, including the odd zeta values. I mention the results that (3) is
irrational [A], that an innity of the odd zeta values are irrational [BR], and that
at least one of (5), (7), . . . , (21) is irrational [R]. The multizeta values have been
interpreted as period integrals, and this interpretation may possibly lead to a better
understanding of them [KZ], [D1]. For more details and references see [V].
526 V. S. VARADARAJAN
3. Divergent series
It was in the systematic theory of innite series (beyond explicit evaluations)
that Euler made one of his greatest contributions, namely his creation of a theory
of divergent series, or at least the rst steps towards such a theory. He knew that by
attempting to associate numbers as values of diverging series he was going beyond
the connes of what people were used to thinking about, and yet he insisted that
one has to build a theory of divergent series in order to free analytical methods
from articial limitations. There is no better place to hear his viewpoint than from
his great 1760 paper (communicated in 1755) De seriebus divergentibus [E1]:
Notable enough, however, are the controversies over the series
1 1 + 1 1 + 1 1 + . . . whose sum was given by Leibniz as
1/2, although others disagree. No one has yet assigned another
value to that sum, and so the controversy turns on the question
whether the series of this type have a certain sum. Understanding
of this question is to be sought in the word sum; this idea, if thus
conceivednamely the sum of a series is said to be that quantity to
which it is brought closer as more terms of the series are takenhas
relevance only for convergent series, and we should in general give
up this idea of sum for divergent series. Wherefore, those who thus
dene a sum cannot be blamed if they claim they are unable to
assign a sum to a series. On the other hand, as series in analy-
sis arise from the expansion of fractions or irrational quantities or
even transcendentals, it will in turn be permissible in calculation
to substitute in place of such a series that quantity out of whose
development it is produced. For this reason, if we employ this def-
inition of sum, that is to say, the sum of a series is that quantity
which generates the series, all doubts with respect to divergent se-
ries vanish and no further controversy remains on this score, in as
much as this denition is applicable equally to convergent or diver-
gent series. Accordingly, Leibniz, without any hesitation, accepted
for the series 1 1 +1 1 +1 1 +. . . , the sum 1/2, which arises
out of the expansion of the fraction 1/1 + 1, and for the series
1 2 +3 4 +5 6 +. . . , the sum 1/4, which arises out of the ex-
pansion of the formula 1/(1+1)
2
. In a similar way a decision for all
divergent series will be reached, where always a closed formula from
whose expansion the series arises should be investigated. However,
it can happen very often that this formula itself is dicult to nd, as
here where the author treats an exceptional example, that divergent
series par excellence 11+26+24120+7205040+. . . , which
is Wallis hypergeometric series, set out with alternating signs; this
series, in whatever formula it nds its origin and however much
this formula is valid, is seen to be determinable by only the deepest
study of higher Analysis. Finally, after various attempts, the au-
thor by a wholly singular method using continued fractions found
that the sum of this series is about 0.596347362123, and in this dec-
imal fraction the error does not aect even the last digit. Then he
proceeds to other similar series of wider application and he explains
how to assign them a sum in the same way, where the word sum
EULER AND HIS WORK ON INFINITE SERIES 527
has that meaning which he has here established and by which all
controversies are cut o. (From the translation of E. J. Barbeau
and P. J. Leah [BL])
The paper shows clearly the fact that he understood what was involved, that
the number we associate to an innite series is a matter of convention, and that a
systematic treatment of this question would be very benecial to the development
of Analysis.
This paper is only a culmination of his ideas on the subject which had been in
gestation for many years prior to its communication. In 1745 he had discussed these
matters in letters to Goldbach and Nicholas Bernoulli, especially the summing of
the factorial series

n0
(1)
n
n!,
which he called the divergent series par excellence. In it Euler has this to say (free
translation from German):
. . . I believe that every series should be assigned a certain value.
However, to account for all the diculties that have been pointed
out in this connection, this value should not be denoted by the name
sum, because usually this word is connected with the notion that a
sum has been obtained by a real summation: this idea however is
not applicable to seriebus divergentibus. . . .
In the nineteenth century, with Abel, Cauchy, Dedekind, and Weierstrass, rig-
orous foundations were laid for Analysis, and divergent series were banned as the
work of the devil. With the rise of these formal principles Eulers reputation also
suered, and people began to misunderstand what he was trying to do and started
thinking of him as a loose mathematician. It should have been clear that no one
who calculated the values of many convergent series to tremendous accuracy would
have loose ideas about when series diverge and what their values are. His true
greatness in these matters was not appreciated till a century afterwards when a
genuine theory of divergent series was created and it became clear how far he was
ahead of his time [Ha], [C].
It is true that Euler had some misconceptions regarding summation of divergent
series. He appeared to believe that all series could be summed by some procedure
or other and also that in general all summation procedures would lead to the same
value. The actual theory of divergent series shows that the situation is much more
subtle. However in my opinion these are dierences in detail that do not alter the
fact that he took the rst steps in creating a true theory of divergent series.
Euler had several dierent methods of summing divergent series, but most of
all he worked with what we now call Abel summation. If

n0
a
n
is a series such
that

n0
a
n
z
n
converges inside the unit disk, we shall say that

n0
a
n
is Abel
summable to the value s if
lim
0<x<1,x1

n0
a
n
x
n
= s.
In particular, if the sum f(z) of the power series extends analytically to a domain
containing z = 1, we can take f(1) as the value of the sum. This is what Euler did.
He often referred to it as the generating function method.
528 V. S. VARADARAJAN
For a large class of divergent series which arise naturally in Analysis this method
turns out to be adequate. Indeed, by using it Euler was able to discover the func-
tional equation of the Riemann zeta function one hundred years before Riemann
did. However for the factorial series this method would fail, as the corresponding
power series has zero radius of convergence. That is why Euler called it the di-
vergent series par excellence. His method of summing it is truly fascinating and
would eventually incarnate into what we now call Borel summation, discovered by
Emile Borel [Bo]. Borel summation and its generalizations have proved surpris-
ingly powerful in many applications, such as quantum eld theory and dynamical
systems.
The functional equation of the zeta. The functional equation of the zeta,
established by Riemann, who was the rst to treat the zeta as a function of the
complex variable s, is given by
(11) (1 s) = 2(2)
s
cos
s
2
(s)(s).
However Euler was led to this equation one hundred years before Riemann. Of
course Euler worked with real s (this was the case even for Dirichlet). Euler veried
this equation exactly for all integer values of s and numerically to great accuracy
for many fractional values as well. For s a positive integer 2, 1 s is a negative
integer and so the series for the corresponding zeta value diverges, showing that
its value has to be interpreted by a summation procedure; naturally Euler used
the generating function method (Abel summation). Of course we now know that
with appropriate growth conditions the verication at the integers establishes the
functional equation. As usual Euler worked with
(s) = 1 2
s
+ 3
s
= = (1 2
1s
)(s)
rather than (s) for better convergence and more accurate numerical evaluation.
For the functional equation is
(12)
(1 s)
(s)
=
2
s
1
2
s1
1
cos
s
2

s
(s).
For s = 2k an even positive integer
(1 2k) = 1
2k1
2
2k1
+ 3
2k1
. . .
with generating function
1
2k1
x 2
2k1
x
2
+ 3
2k1
x
3
. . . .
We now write x = e
y
so that x 1 corresponds to y 0+. From
e
y
e
2y
+e
3y
=
1
e
y
+ 1
we get, on dierentiating repeatedly,
1
m
e
y
2
m
e
2y
+ 3
m
e
3y
= (1)
m
d
m
dy
m
_
1
e
y
+ 1
_
.
Since
1
e
y
+ 1
=
1
2
+

k=1
(1 2
2k
)B
2k
(2k)!
y
2k1
EULER AND HIS WORK ON INFINITE SERIES 529
we have
(1)
m
d
m
dy
m
_
1
e
y
+ 1
_

y=0
=

1
2
if m = 0
0 if m = 2, 4, 6, . . .
2
2k
1
2k
B
2k
if m = 2k 1.
Thus Euler obtained his remarkable formulae
1
m
2
m
+ 3
m
=

1
2
if m = 0
0 if m = 2, 4, 6, . . .
2
2k
1
2k
B
2k
if m = 2k 1.
In particular
(1 2k) =
2
2k
1
2k
B
2k
giving
(13)
(1 2k)
(2k)
=
2
2k
1
2
2k1
1
cos k

2k
(2k 1)!,
which is (12) for s = 2k. Here we have used the even zeta values computed already
by him. For s odd (12), after bringing (s) to the right side, is trivially true since
both sides are 0. This is in fact the reason why Euler substituted cos k for (1)
k
in the previous calculation, clearly suggesting that he was thinking of the functional
equation for non-integral values of s. Euler then proceeded to write that . . . I shall
hazard the conjecture that the relation (12) is true for all s. . . .
He then proceeded to verify this conjecture as far as he could for other values of
s. For s = 1
(1) = log 2, (0) =
1
2
, (0) = 1 1 + 1 1 + 1 1 +. . . .
The Leibniz series has the sum 1/2. For s = 0 the calculations are the same since
the roles of s and 1 s are reversed. Euler notes explicitly that the transformation
s 1 s leaves the functional equation invariant, a fact that is easily veried and
depends on the relation
(s)(1 s) =

sin s
,
which he knew. Thus the functional equation was veried whenever s is any integer,
positive, negative, or zero.
For fractional values of s Euler resorted to numerical evaluation. For s = 1/2 we
have (s) = (1 s) = (1/2), and the verication of (10) depends on the relation

_
1
2
_
=

,
which Euler knew. Euler treated s = 3/2, and it appears that he had veried the
functional equation for s = (2i + 1)/2 (i = 1, 2, 3, 4, 5, 6, . . . ). For s = 3/2 Euler
used the Euler-Maclaurin sum formula to compute the sums appearing on both sides
of (11). This of course is not the same thing as being summed by the generating
function method, but Euler went ahead because he rmly believed that all methods
of summation lead to the same value. (The Euler-Maclaurin summation method
was resurrected by Ramanujan centuries later.) Much later Landau justied Eulers
method of summation for all s.
530 V. S. VARADARAJAN
The summation of the factorial series. The summation of the factorial series,
the divergent series par excellence as he called it, presented serious problems for
Euler and his philosophy of using the generating function for summing divergent
series. The formal power series
f(x) = 1 1!x + 2!x
2
=

n0
(1)
n
n!x
n
does not converge anywhere and so his generating function has only a formal ex-
istence. Euler [E1] introduced the function g(x) = xf(x), and his method starts
with the dierential equation satised by g:
x
2
dg
dx
+g = x,
for which one can nd the complete integral
g(x) = e
1/x
_
1
x
e
1/x
dx +ce
1/x
,
c being an arbitrary constant. It is very interesting to note that the function e
1/x
is rapidly decreasing as x 0+ and its derivatives at 0 are all 0; it is in fact the
classical Cauchy function, which is at at 0. The appearance of at functions is
characteristic of the theory that Euler discovered in the process of his summing the
factorial series. It is easy to see that as x 0+,
e
1/x
_
x
0
1
t
e
1/t
dt = O(x)
and so, by choosing c = 0, we match the asymptotics of the integral with that of
the formal series to get, with Euler,
(14) f(x) =
1
x
e
1/x
_
x
0
1
t
e
1/t
dt
so that
(15) f(1) =
_
1
0
e
11/t
t
dt,
which reduces the problem to the evaluation of the integral. Today we would not
write equality in (14) and (15), but use some notation such as to indicate that
the two sides are asymptotic. Euler rst evaluates the integral numerically; writing
h for the integrand and using the trapezoidal rule for approximating the integral
by
1
10
_
(1/2)h(0) +h(1/10) +h(2/10) +. . . h(9/10) + (1/2)h(1)
_
,
he gets
1 1! + 2! 3! + = 0.59637255 . . . .
This is a surprisingly accurate evaluation, which may be partially explained by
using the Euler-Maclaurin summation formula (see below).
He was not satised with this and wanted a more accurate evaluation of the
integral. For this purpose he discovered a continued fraction for it,
f(x) =

n0
(1)
n
n!x
n
=
1
1+
x
1+
x
1+
2x
1+
2x
1+
3x
1+
3x
1+
etc.
EULER AND HIS WORK ON INFINITE SERIES 531
Using this he nds for the integral the value
I = 0.596347362123 . . . (I = 0.5963473625 using MAPLE).
Euler did not stop with this single example. In his letter to Nicolaus Bernoulli
that I mentioned he says that similar methods can be applied to many other series
and related them to corresponding continued fractions. He treated in [E1] for
instance the class of series
g = g
mpq
= x
m
px
m+q
+p(p +q)x
m+2q
p(p +q)(p + 2q)x
m+3q
. . . .
For m = p = q = 1 one obtains the factorial series xf = g
111
. g satises
x
q+1
g

+ [(p m)x
q
+ 1]g = x
m
so that (matching asymptotics at x = 0 as before)
g e
1
qx
q
x
mp
_
x
0
e
1/qt
q
t
pq1
dt.
Euler nds for g the continued fraction
g
mn
=
x
m
1+
px
q
1+
qx
q
1+
(p +q)x
q
1+
2qx
q
1+
(p + 2q)x
q
1+
3qx
q
1+
(p + 3qx
q
)
1+
etc.
If we take p = m = 1 and q = 2, we get the series
g = x 1x
3
+ 1.3x
5
1.3.5x
7
+ 1.3.5.7x
9
. . . .
For g the integral is given by
g e
1
2x
2
_
x
0
e
1/2t
2 1
t
2
dt.
Thus he nds
1 1 + 1.3 1.3.5 + 1.3.5.7 =
_
1
0
e
[1(1/t
2
)]/2
1
t
2
dt
as well as the continued fraction
1
1+
1
1+
2
1+
3
1+
4
1+
5
1+
etc.
Using the same methods as before he nds for this series the value
0.65568 (0.6556795424 using MAPLE).
It is important to note that these continued fractions of Euler are quite dierent
from the classical simple continued fractions which are of the form
a
0
+
1
a
1
+
1
a
2
+
. . .
where the as are positive integers. The Eulerian ones are of the form
1
1+
a
1
1+
a
2
1+
. . .
where the a
i
are positive integers. In his treatment Euler implicitly assumes that
these behave like the simple ones. It can be shown that Eulers assumption is in
fact true and that these continued fractions do converge, at least when a
n
= O(n),
the partial convergents being alternately above and below the true value. The
nature of the values of these continued fractions (algebraic, transcendental) is not
known. Calculating with Maple, we nd that the limiting value of the continued
fractions of Euler is accurate to a huge number of decimal places. Restricted to
532 V. S. VARADARAJAN
hand computing, Euler resorted to a method of calculating which produced a result
accurate to about 8 decimal places. To explain his method let us think of the
nite continued fractions as the result of applying a sequence of fractional linear
transformations. Thus
1
1+
a
1
1+
a
2
1+
. . .
a
n1
1+
a
n
1 +t
= F
a
1
F
a
2
. . . F
a
n
(t) =: H
n
(t)
where F
a
is the map z a/1 +z. Eulers idea was to evaluate, for a moderately
large N, H
N
(t) = F
1
F
2
. . . F
N
(t) not at t = 0 but at a suitable point t = t
N
, a
step that is justiable in view of the following result that can be proved: when
x
n
= O(n), for any arbitrary sequence (t
n
), t
n
[0, ], the sequence H
n
(t
n
) has a
limit as n , and this limit is independent of the sequence, equal for example
to the limit when we choose t
n
= 0 for all n. The idea behind evaluating H
N
(t)
at t = t
N
instead of t = 0 is to improve the rapidity of convergence to the limit.
Euler starts with the choice of t
N
as the xed point of F
N+1
. For N = 20 this gave
for Euler a value accurate to about 4 decimals, but then, remarkably, he makes a
further variation in the choice of t
N
that gets him to about 8 decimal place accuracy
(see [BL]). One can explain his method and even improve on it, as was shown to
me by Deligne [D], but it would take me too far aeld to do it here. I hope however
to discuss these matters in more detail on a later occasion [DV].
In the theory of summability of divergent series that grew in the nineteenth
century, it is the theory of Borel summation that allows us to put Eulers ideas on
the summation of factorial-like series into proper perspective. The Borel transform
f

of the formal power series f =

n0
a
n
z
n
is the formal power series
f

(z) =

n0
a
n
z
n
n!
.
At a completely formal level we can then recover the original series by the formula

n0
a
n
=
_

0
e
t
f

(t)dt.
If the a
n
= O((n!)

), for < 1, then f

is entire, but not if 1. In Borels

theory, the right side, whenever it makes sense, is dened as the Borel sum of the
series on the left (if the series is absolutely convergent, the Borel sum coincides
with the ordinary sum). For the Euler series, f

(w) = (1 +w)
1
has a singularity
at only w = 1 and so the Borel sum makes sense. We then have

n0
(1)
n
n!z
n

_

0
e
w
1 +wz
dw =: F(z)
in the sense of Borel. The function F(z) dened by the right side is analytic for
z = re
i
with || < and is asymptotic to the formal power series on the left in
the Poincare sense:

F(z)

0mn
(1)
m
m!z
m

= O(|z|
n+1
) uniformly in || )
for each > 0. Actually, the asymptotics holds in a sharper sense:

F(z)

0mn
(1)
m
m!z
m

(n + 1)!|z|
n+1
for all || , n 0)
EULER AND HIS WORK ON INFINITE SERIES 533
where K

> 0 is a constant independent of z and n; this is a renement of the

Poincare estimate that gives the dependence on n also. It is referred to as strong
asymptotics; one is allowed an additional
n
for the estimate of the remainder. It
follows from a famous theorem of Watson that F is uniquely determined by the
factorial series as the only analytic function in the sector || < strongly
asymptotic to it. With suitable modications one can treat all of Eulers examples
in an analogous manner.
The integral discussed above,
_

0
e
w
1 +zw
dw
is the inverse Borel transform of (1 + zw)
1
, which is the Borel transform of the
factorial series

n0
(1)
n
n!z
n
. It is easy to see by a change of variables that it is
the integral that Euler wrote. Indeed,
_

0
e
w
1 +xw
dw =
1
x
e
1/x
_

0
e
1/t
t
dt (w = 1/t 1/x) .
Similarly, the series g
mpq
has the Borel transform

n0
(1)
n
x
m+rq
p(p + q)(p + 2q) . . . (p + (r 1)q)
n!
w
r
= x
m
1
(1 +qx
q
w)
p/q
so that the Borel sum of g
mpq
is the integral
g
mpq
x
m
_

0
e
w
(1 +qx
q
w)
p/q
dw.
By the substitution w = 1/qt
q
1/qx
q
this integral goes over to
e
1
qx
q
x
mp
_
x
0
e
1/qt
q
t
pq1
dt,
which is the integral obtained by Euler for this series. Since both the series g
mpq
and the associated integral depend on x through x
q
, it is better to work in the
-plane where = z
q
. Then the integral
_

0
e
w
(1 +qw)
r
dw (r > 0)
is holomorphic on the cut plane with | arg()| < , strongly asymptotic to

n0
(1)
n
r(r + 1) . . . (r +n 1)
n!

n
on sectors | arg()| for each > 0. I omit the details (see however [Bo],
[Ca]).
In my opinion, the understanding of these ultradivergent series as asymptotic
expansions of solutions of dierential equations having an integral representation,
which is one of the many features of the work of Euler on the summation of such
series, was a fantastic discovery. It would be more than a century before Poincare
[P] and Emile Borel [Bo] rediscovered this theme as a part of the theory of linear
meromorphic dierential equations with irregular singularities and the asymptotics
of their solutions. For these, unlike the Frobenius series when the singularity is
regular, the formal solutions do not converge, although one can construct a full
fundamental matrix of formal solutions. On any sector with vertex at one of the
534 V. S. VARADARAJAN
singularities, one can construct an analytic solution (germ of it actually), asymptotic
to the formal matrix, in the sense that Poincare introduced. Note that there are at
matrices of solutions in general and so the analytic solution is not unique, although
summability methods allow one to pick the solutions in a canonical manner. In the
modern versions of the theory these sheaves of at functions, or rather their rst
cohomologies, form an obstruction to linking the formal theory with its analytic
counterpart, and once they are taken into account, one can obtain a satisfactory
theory [V1]. These summability methods have had great impact in many areas.
This is however only one aspect of the theme that Euler started with his summation
of the ultra divergent series of factorial type. The issues concerning the structure
and numerical evaluations of the continued fractions of the Euler type, as well as
their relation to the solutions of the dierential equations, are still not fully resolved.
One last remark is appropriate before we wind up this brief discussion of diver-
gent series and their summation. In modern analysis, especially in the theory of
Fourier series and representation theory, another method of summation is widely
used, the so-called smeared summation. Unlike most summation methods touched
upon above, this method makes sense in all dimensions and has great exibility. If
{f
n
(x)} is a sequence of functions dened on R
n
or the torus T
n
or, more generally,
on a Lie group G, then

n
f
n
(x)
is summable in the weak sense if for all smooth functions with compact support ,
the sum

n
_
G
f
n
(x)(x)dx = L()
is convergent; the limit L is then a distribution in the sense of L. Schwartz. Very
often L is dened by a function L(x), namely
L() =
_
G
L(x)(x)dx,
and then we say that

n
f
n
(x) = L(x)
in the sense of distributions. One is led to think of this as smeared summation
because in many practical problems, the f
n
represent some physical quantities,
and the actual value of f
n
(x) at a space time point is very dicult to determine
exactly; only a space time average of f
n
around x is measurable. One says that
the measurement of f
n
(x) is smeared. This method was used by Harish-Chandra
to dene the characters of innite dimensional irreducible unitary representations.
Some of the main results of Fourier analysis on G can be viewed as a formula
expressing the Dirac delta at the identity as a linear combination of the characters
of the irreducible unitary representations. The simplest of these is on the circle
where the formula is
=

n=
e
inx
.
Euler was interested in the sum on the right side. He proved, using his favorite
Abel summation, that the right side is 0 except when x 0 mod 2. It is easy to
EULER AND HIS WORK ON INFINITE SERIES 535
see that if
s
N
(x) =
N

n=N
e
inx
,
N
(x) =
s
0
(x) +s
1
(x) +. . . s
N
(x)
N + 1
,
then

N
(x) 0 (N )
when x 0 mod 2 while

N
(0) = N + 1,
showing that the limit should be thought of in the smeared sense and is . A proof
is quite easy, using elementary Fourier analysis.
4. Summation formula
Throughout his life Euler was a tireless calculator, delighting in numerical cal-
culations in almost all areas he worked in: astronomy, mechanics, innite series,
and so on. In the theory of innite series he calculated the zeta values to a huge
number of decimal places, using what we now call the Euler-Maclaurin summation
formula. This was a favourite tool of his, and he used it in many ingenious ways to
evaluate the results in many problems with great numerical accuracy. For instance
he calculated to a huge number of decimal places what we now call Eulers constant,
the summation value of the factorial series, the numerical value of = 3.14 . . . , and
so on.
The summation formula was read by Euler before the St. Petersburg Academy
in 1734. Maclaurin discovered it independently around 1740. There was however
no argument about priority; Euler just contented himself, in response to enquiries
from Stirling, with the remark that the result is known and was read before the
Academy in 1734. This was characteristic of him and his generous ways of accepting
what others had done.
Let us recall the formula. If f is a nice function, the summation formula asserts
that
1
2
f(0) + f(1) +f(2)+ +f(N 1) +
1
2
f(N) =
_
N
0
f(x)dx
+

k=2
B
k
(k)!
[f
(k1)
(N) f
(k1)(0)
]
where the B
k
are the Bernoulli numbers dened by
1
1 e
z
=
1
z
+
1
2
+

k=2
B
k
k!
z
k1
.
One has, with the convention B
0
= 1,
B
1
=
1
2
, B
k
= 0 (k 2, k odd), B
2
=
1
6
, B
4
=
1
30
, . . . .
Formally one wants to nd S(x) such that f(x) = S(x) S(x 1); if D = d/dx,
then the above equation can be written as (1 e
D
)S = f or f = (1 e
D
)
1
f,
from which, expanding in powers of D, we get the result. It is possible to impose
conditions on f that make it clear that this is an asymptotic expansion: the error
committed is at most the size of the last term taken into account. Euler knew of
536 V. S. VARADARAJAN
this because he always maintained that one has to stop when the terms of the series
begin to diverge.
There is another way to understand this formula that links it with Fourier analy-
sis and Poisson summation formula. I owe it (as well as the illustrative treatment of
the Euler evaluation of the factorial series that follows as an application) to Pierre
Deligne [D] and I wish to describe it briey.
Let f be a function with compact support and of bounded variation and let its
Fourier transform

f be dened by

f(y) =
_
f(x)e
2ixy
dx.
Then we have (the Fourier integral analogue of Dirichlets theorem for Fourier
series)
1
2
[f(x+) +f(x)] = lim
R
_
R
R

f(y)e
2ixy
dy.
Let us now dene the function on R/Z (with the natural map x x)

f( x) =

nZ
f(x +n).
Then

f is of bounded variation and its Fourier coecients are {

f(m)}
mZ
. By
Dirichlets theorem we have
1
2
[

f(0+) +

f(0)] =

mZ

f(m),
which is the Poisson summation formula in this context; the series on the right
is in general not absolutely convergent, and one has to interpret it as a principal
value, i.e., as lim
M

M
m=M

f(m). If f is in addition continuous on [0, N] and

vanishes outside [0, N], then
1
2
[

f(0+) +

f(0)] =
1
2
f(0) +f(1) + + f(N 1) +
1
2
f(N)
so that the above formula becomes
1
2
f(0) +f(1) + +f(N 1) +
1
2
f(N) =
_
f(x)dx +

m=0

f(m).
It is now a question of determining the asymptotic behaviour of the terms

f(m)
that appear on the right side. This is controlled by the singularities of f and its
derivatives at 0 and N.
Assume now that f is actually C

on the closed interval [0, N] and vanishes

outside [0, N]. Then, repeated integration by parts yields the asymptotics

f(m)

k=1
f
(k1)
(0) f
(k1)
(N)
(2im)
k
(m ).
So

m=0

f(m)

m=0

k=1
f
(k1)
(0) f
(k1)
(N)
(2im)
k
(m ).
EULER AND HIS WORK ON INFINITE SERIES 537
We now interchange the order of summation of course. The terms for odd k for m
and m cancel, while for even k = 2r the summation over m gives, using Eulers
formulae (7) for (2r),
(1)
r
(2)
2r
2(2r) = (1)
r
(2)
2r
(1)
r1
2
2r
B
2r
(2r)!

2r
=
B
2r
(2r)!
.
This gives the summation formula. If one does not want to use the full asymptotic
expansion for

f(m), one can stop at the term k = 2r + 2 with a remainder term
1
(2im)
2r+2
_
N
0
f
(2r+2)
(x)e
2ixy
dx,
which is majorized by
||f
(2r+2)
||
1
(2m)
(2r+2)
where || ||
1
is the L
1
-norm. We then see that the error in stopping at the (2r)
th
term in the summation formula is

B
2r+2
(2r + 2)!

_
|f
(2r+1)
(N) f
(2r+1)
(0)| +||f
(2r+2)
||
1
_
.
If we assume (as Hardy does in his treatment [Ha]) that f
(2r+2)
> 0 in (0, N), we
can replace the L
1
-norm by
|f
(2r+1)
(N) f
(2r+1)
(0)|
so that the error is
2

B
2r+2
(2r + 2)!
_
f
(2r+1)
(N) f
(2r+1)
(0)
_

.
When one studies the convergence or summability of the Euler-Maclaurin series or is
concerned with numerical evaluations, it is thus a question of balancing the growth
of the derivatives of f with that of the Bernoulli numbers. Euler was extremely
skillful in such calculations. One should also remark that the expansion can equally
be used to approximate the integral by the sum. Finally, the two features of the
Euler formula, namely, the weights 1/2 for the boundary terms and the Bernoulli
numbers, enter Delignes treatment for reasons dierent from the ones in Eulers
original treatment: the boundary weights 1/2 come from Dirichlets theorem, and
the Bernoulli numbers from the zeta values (see also [Ha], p. 330).
As an illustration let us use the summation formula to approximate
_
1
0
h(t)dt h(t) =
e
1(1/t)
t
by

_
1
2
h(0) +h() +h(2) + +h(9) +
1
2
h(1)
_ _
=
1
10
_
,
which Euler did in his rst evaluation of the sum of the factorial series. To under-
stand the surprising accuracy of this rst evaluation we use the summation formula
(applied to f(t) = h(t) (0 t 1/)) to write the integral as a sum plus error
terms. Now h is at at 0 and h

(1) = 0 and so the B

2
term is 0. If we take the
approximation through the term B
2
/2!, the error committed is comparable to the
rst term omitted in the summation formula, namely to the size of the B
4
term.
538 V. S. VARADARAJAN
Now a simple calculation shows that h
(3)
(1) = 4. Hence the absolute value of the
B
4
term is
|B
4
|
4!
1
10
4
|h
(3)
(1)| =
1
180 10
4

1
10
6
,
which is the rough estimate of the error in Eulers approximation, giving at least a
partial explanation as to why his rst evaluation was so accurate.
5. Concluding remarks
I have not touched on many aspects of Eulers work of great current interest.
For instance, in number theory he was indeed the great pioneer. He developed
the foundations of the subject so that he could obtain the proofs of most of the
assertions of Fermat, especially on problems involving the sums of two squares. He
generalized the whole set up to one that asked what primes could be written in the
form x
2
+Ny
2
for composite N, which led him to quadratic reciprocity and beyond;
indeed, to survey his work from the modern point of view requires application of
such sophisticated theories as class eld theory. He discovered the product formula
for the zeta function
(s) =

p
1
1
1
p
s
,
which he would write as
1 +
1
2
s
+
1
3
s
+ =
2
s
.3
s
.5
s
.7
s
. . .
(2
s
1)(3
s
1)(5
s
1)(7
s
1) . . .
as well as the product formula for a few twisted series (with real characters mod 4,
6, 8, 12)
1 +
(2)
2
s
+
(3)
3
s
+ =

p
1
1
(p)
p
s
.
From these he deduced, using their behaviour at s = 1, for example that there
are innitely many primes in each residue class p 1(mod 4) and some classes
mod 8. Ever since, such products over primes have been called Euler products.
Almost a hundred years after Eulers discoveries Dirichlet would take up this theme
and advance it spectacularly, associating Euler products to all characters, real or
complex, for any modulus N, using their behaviour at s = 1 to conclude that there
are innitely many primes in each residue class mod N prime to N for all moduli
N. The further history of these ideas at the hands of modern masters such as Artin,
Weil, Ramanujan, Deligne, Langlands, and a host of others is too well known to
bear repetition here.
Inspired by Fagnanos work which came to his attention while he was at the
Berlin Academy, Euler discovered the addition formula for elliptic integrals, which
is the forerunner for the modern theory of elliptic curves and abelian varieties. It
is of course impossible to go into detail about all of these. It is my hope that my
discussion is enough to rekindle interest in Eulers work and to inspire younger
people to look at his work with new eyes and focus.
About the author
V. S. Varadarajan is a professor of mathematics at the University of California
at Los Angeles. His research interests include representations of Lie and super
EULER AND HIS WORK ON INFINITE SERIES 539
Lie groups. His most recent books are Euler Through Time: A New Look at Old
Themes and Super Symmetry for Mathematicians: An Introduction.
References
[A] R. Apery, Irrationalite de (2) et (3), Asterisque 61 (1979), 11-13.
[Bo] E. Borel, Le cons sur les Series Divergentes,

Editions Jacques Gambay, 1988 (Reprinting of
the original 1928 work).
[BL] E. J. Barbeau, and P. J. Leah, Eulers 1760 paper on divergent series, Historia Mathematica,
3(1976), 141160. MR0504847 (58:21162a)
[BR] K. Ball and T. Rivoal, Irrationalite dune innite de valeurs de la fonction zeta aux entiers
impairs, Invent. Math. 146(2001), no. 1, 193207. MR1859021 (2003a:11086)
[C] P. Cartier, Mathemagics (A Tribute to L. Euler and R. Feynman), Lecture Notes in Phys.,
vol. 550, Springer, Berlin, 2000, 667. MR1861978 (2003c:81002)
[Ca] B. Candelpergher, J. C. Nosmas, and F. Pham, Approche de la resurgence, Hermann, Paris,
1993. MR1250603 (95e:34005)
[D] P. Deligne, Letters (Personal communication).
[D1] P. Deligne, Lectures at UCLA, Spring 2005.
[DV] T. Digernes and V. S. Varadarajan, Notes on Eulers continued fractions (In preparation).
[E] L. Euler, Omnia Opera. All of Eulers papers in pure mathematics are in Series I of his
Omnia Opera. For most of the questions discussed here, see I-8, I-14, I-15.
[E1] L. Euler, De seriebus divergentibus, Opera Omnia, I, 14, 585617.
[Ha] G. H. Hardy, Divergent series, Oxford, 1973. MR0030620 (11:25a)
[Hi] Omar Hijab, Introduction to Calculus and Classical Analysis, Springer, 2007 (Second Edi-
tion). MR1449395 (98e:26001)
[KZ] M. Kontsevich and D. Zagier, Periods, in 2001 and Beyond, 771808, Springer, 2001.
MR1852188 (2002i:11002)
[P] H. Poincare, Sur les integrales des equations lineaires, Acta Math. 8(1886), 295344; Oeu-
vres, TI, Gauthier-Villars (1928), 290332. MR1554701
[R] T. Rivoal, Irrationalite dau moins un des neuf nombres (5), (7), . . . , (21), Acta Arith.
103 (2002), no. 2, 157167. MR1904870 (2003b:11068)
[V] V. S. Varadarajan, Euler Through Time: A New Look at Old Themes, AMS, 2006. This is
a partial source for almost all the matters discussed and the references for works mentioned
in this article. MR2219954
[V1] V. S. Varadarajan, Linear meromorphic dierential equations: a modern point of view.
Bull. Amer. Math. Soc. (N.S.) 33 (1996), no. 1, 142. MR1339809 (96h:34011)
Department of Mathematics, University of California, Los Angeles, Los Angeles,
California 90095-1555
E-mail address: vsv@math.ucla.edu